materials-discovery-and-bulk-selection

name	materials-discovery-and-bulk-selection
description	Use this skill for materials discovery and bulk structure selection before slab construction, including database query strategy, candidate filtering, and export readiness.

materials-discovery-and-bulk-selection

Overview

Use this skill to turn an open-ended catalyst search into a shortlist of downloaded bulk structures with traceable Materials Project provenance.

ExperimentSpecialist may use the MP lookup tools directly for lightweight database retrieval. When this skill is invoked inside a delegated materials workflow, perform the lookup here and report precise API-key, client-package, criteria, or field blockers rather than saying materials discovery is generally unavailable.

Quick Start

1. Define search criteria and requested fields before querying Materials Project.
2. Choose a non-empty criteria, non-empty fields, and an output_csv_rel ledger path before calling mp_search_materials.
3. Use mp_search_materials to write a CSV candidate table, not just a chat summary.
4. Prune the candidate table before downloading structures.
5. Use mp_download_structure only for the shortlisted mp_id values.

Allowed tools

• mp_search_materials
• mp_download_structure

Workflow

1. Search with explicit criteria

• mp_search_materials fails on empty criteria or empty fields.
• The tool writes the full result table to output_csv_rel and returns preview rows for quick inspection.
• Keep the requested fields aligned with the intended downstream ranking so the CSV is directly usable.
• Common criteria keys include material_ids, elements, chemsys, formula, is_stable, energy_above_hull, formation_energy, band_gap, crystal_system, spacegroup_number, spacegroup_symbol, num_sites, density, volume, and num_elements.
• Range criteria should be passed as [min, max] pairs.
• Common output fields include material_id, formula_pretty, energy_above_hull, formation_energy_per_atom, band_gap, crystal_system, spacegroup_number, spacegroup_symbol, nsites, density, and volume.

2. Prune before download

• Use the CSV output as the candidate ledger instead of retyping IDs from memory.
• Keep the filter logic explicit so later bulk or slab comparisons remain auditable.

3. Download only retained candidates

• mp_download_structure writes conventional standard structures under output_dir.
• It supports partial success: some mp_id values can fail while others download cleanly.
• Use the returned results and errors rather than assuming every requested ID resolved.

4. Hand off the shortlist

• Once the bulk shortlist is stable, hand it to bulk-relax-and-reference for relaxation and reference-state cleanup.

Method-critical defaults

• Keep the search criteria, requested fields, and shortlist filter fixed while you compare candidates.
• Do not treat a candidate table built with a different field set as interchangeable with the current shortlist.

Output Contract

Return:

• search CSV path
• retained mp_id list
• downloaded structure path(s)
• any partial-download failures that still need follow-up

References

• If a task needs advanced MP field selection, inspect the tool schema before broadening the search request.
• Use bulk-relax-and-reference when the discovery phase is finished and you need a real reference structure.